Improving 5 G base station placement through precise rooftop detection using super-resolution diffusion models and satellite image analysis

Abstract The accurate deployment of 5 G base stations (BSs) in urban environments is essential for achieving optimal network performance. In these scenarios, the most common positions for installing BSs are rooftops, which, however, given the complex topography and diverse building structures, present significant challenges when identifying suitable locations. This paper proposes an enhanced method for rooftop detection, integrating diffusion models based on super-resolution with segmentation using convolutional neural networks. Starting from the input image, a super-resolution model is applied to generate sliding windows on which re-inference is performed, thereby improving both the resolution and prediction accuracy for this type of object. By refining these detections, the placement of 5 G base stations is undertaken in a practical, industrial way, thus allowing network operators to perform a more real-world network optimization. The results demonstrate a significant improvement in detection accuracy, directly contributing to more efficient 5 G base station deployment in densely populated urban areas. This methodology offers a scalable, adaptable, and effective solution based on the context of the images it applies to.